基于日志的流程相似度计算方法适用性分析

Applicability analysis of log-based process similarity computing method

针对现有业务流程相似度计算方法的不足,从流程日志出发计算流程相似度,并根据以流程日志为输入的特点对现有方法进行了改进,提出两种改进方法。一种方法是通过日志重构流程模型,基于模型结构计算流程相似度,但是为体现流程日志中对流程结构某些执行路径或分支的倾向性,在流程模型上重演日志轨迹,利用A*算法寻找匹配日志轨迹的最佳重演方式,识别最佳重演方式下被执行的流关系,以对流关系加权生成加权BPMN模型计算流程相似度;另一种方法不需重构模型,基于任务序列计算流程相似度,为支持任务相似度计算,定义了任务间的距离,由此得到支持任务距离度量的轨迹距离,进一步利用轨迹多集计算流程相似度。通过在不同的真实数据集上的实验,分析了两种方法在相似度计算结果的异同之处,并总结了两种方法的适用范围。

To address the issues of existing business process similarity computation methods,the process similarity based on process logs was computed,these existing methods were improved based on the characteristics of process logs as input and two improved methods were proposed.One method reconstructed the process model through the log and computed the process similarity based on the model structure.However,to embody the tendency of some execution path or branch in the process log,the log trajectory was replayed on the process model,the best reacting model was used to find the matching log track by A*algorithm,and the executed flow relationship was identified under the best reacting model.On this basis,the weighted BPMN model generated by flow relationship which was weighted was used to compute the process similarity.The other method did not need to reconstruct the model and compute the process similarity based on the log behavior sequence.To support the task similarity computing,the distance between tasks was defined.Thus,the distance of task sequence supporting task distance was obtained,and the business process similarity was computed by using the multiple set of task sequence.Through the experiments on different real data sets,the similarities and differences of the two methods in similarity computing were analyzed,and the scope of application of the two methods was summarized.